Jonathan, first of all, we're talking about replication jobs, right?
Basically, if direct mode is used, data is being transferred between proxy servers. In case of WAN accelerated transfer, the data flow is the following
: source storage -> source proxy -> source WAN -> target WAN -> target proxy -> target datastore. So, in case proxy and WAN accelerator are deployed on different servers, there's another leg for data to pass.
That said, what is the link you have between locations and what is the current bottleneck reported by the replication job in question? Depending on that, you may or may not need WAN accelerators to be involved.